I had the delight to have a few hours on a Sunday afternoon just to myself, so I could freely indulge in going down a rabbit hole of reading, getting inspired and experimenting. Some of the journey was so nice that later I decided to write it up and share it. This might become a series of public diaries so anyone else can join in these tumbles down the 🐇🕳️
It all started with the existential dread of 170-something open tabs in Firefox. I finally decided to try to do something about it. I actually have a vertical tab bar inspired by Arc browser using the extension Tree Style Tab that automatically groups tabs under "parent" tabs you open "child" tabs from, so you have an idea which tabs belong together. I was wondering if it's possible to create a blank tab that only serves as a parent so that I could group together large number of tabs that are all research for a project or similar. As it turns out it is, but not through the UI, instead by typing in the address bar: about:treestyletab-group?title=LlamasAndWizards
LLMs, innovation and capitalism
Having got rid of the clutter, I could finally rediscover some articles I had opened weeks ago, but never got around to read. One of these was LLMs, Innovation, and Capitalism by Belinda Li highlighting some important issues facing researchers working with Large Language Models.
First, not being able to study models that are only accessible through restricted APIs:
Research that requires access to model logits and weights, such as interpretability studies, model editing, and controlled generation studies (which allow us to understand and control the decisions made by these models), cannot be done except on much smaller-scale language models, which can display very different behavior compared to large language models. Thus, a few powerful entities get to close off entire research avenues for the larger community by deciding what forms of research can be conducted and by whom.
Then, the curious case of large corporations only caring about safety and ethics as long as it also aligns with their ability to profit from closed models:
Rather than suggest we stop research in scaling, I believe we should try and curb the worst elements of capitalism that have weaved its way into the field. We already know that private corporations aren’t incentivized to do so—most evidently, in the ousting of Google AI ethics researchers in 2020. Compare this to OpenAI’s safety policy, or corporate legal teams which flag and review each paper before they are released. It just so happens that the latter group is constructed to protect the interest of these corporations. Risks and ethics are important to consider only to the degree that they bring in profits and evade lawsuits; when it’s no longer profitable to consider them, these things are quickly pushed to the wayside.
And why trying to recreate models as open source won't solve the underlying issue of misaligned incentives created by capitalism:
Finally, I believe achieving this future is inherently a political question—a question of how we can reshift decision-making power about who gets access to technologies to the democratic majority. There is clearly a lot of discontent (and existential dread) from NLP researchers about the recent trend towards closed science. But other than lamenting on Twitter and in private discussions, very few concrete actions have been taken. One solution, to simply build open-source replicas, is potentially useful in the short term but does not address the root of the issue—corporations always have an upper hand when they get to utilize open-source collective efforts, but get to privatize and profit off of their own ideas. One may also be tempted to simply put forth a sound rhetorical argument for open-sourcing, though it seems difficult to simply persuade a company against pursuing profit in a capitalist society.
But the version posted in Reboot magazine also had some further links after the main article, amongst others to an article about The soft power of Google Doc publishing in the Escape the algorithm publication. It highlights the huge dark web of amazing resources, published from Google Docs:
Axiom 4: If you build a tool with the ability to publish, so help them god, people will publish
They will publish often, zealously, and without regard for the intended purpose of the tool. Yelp reviews will be co-opted to publish blog posts; Venmo payments will be co-opted to publish poems; spreadsheets will be co-opted to publish personal websites; maps will be co-opted to publish magazines. The arc of specialized publishing bends towards generalized publishing.
Axiom 5: A doc is a distinct, shareable object
A doc teaches you how to get into competitive Pokemon. A doc helps you determine whether you are lesbian. A doc exhaustively screenshots and analyzes food in the 2019 TV series Ming Dynasty. A doc walks you through the steps to rip a CD. A doc is a list of galleries representing only white artists. A doc is a room you need to escape. A doc is a recipe book shared between friends. A doc is a log of dose trips. A doc is the salaries of TV writers. A doc tracks the points awarded on every season of Drag Race. A doc chronicles the activities of an internet account called "Dad." A doc indexes the lore of Starset albums. A doc is a syllabus for decentering whiteness in design history. A doc asks what we should do before January. A doc is a love letter to a friendship. A doc is a toolkit for getting through the pandemic. A doc is a map. A doc is a playlist of experimental films and videos. A doc offers ideas for how to teach ceramics making virtually. A doc sends masks to prisons. A doc is email correspondences between My Little Pony characters. A doc is a place to put texts you want to send to your ex. A doc is a list of things to do after prom. A doc is a poetry mixtape.
Axiom 6: The cheaper and easier a publishing tool is to use, the more ripe it is for challenging power hierarchies
A tool that requires more technical literacy, time, or money becomes inherently inaccessible to many with less power, and its form will appropriately and intentionally project class, prestige, and rarity. For that reason, a feature-rich custom website designed and built by a web development agency can only have so much radical potential. A tool that is free, and whose interface is a common word processor immediately understandable by most creators and consumers, will often incorporate fringe ideas and language in a way that these websites do not.
I haven't got time to check out all those docs yet, but the playlist of experimental films and videos caught my eye and it did not disappoint. I checked out a few, but this in particular got me inspired:
Animated leaves, fern, and water reed. Shot on a stills camera using bulk loaded, Kodak 5205 (motion picture film) and hand processed using Jobo C41 press kit. The stills camera shot down the lens of a modified slide projector, which held, and magnified the loosely placed foliage.
Sound is comprised of field recording, and a Tibetan singing bowl, recorded to a Nagra IV, and stretched / layered digitally.
I just loved those leaves with light shining through them, so decided to experiment a bit myself.
First, I wanted to see how would taking a photo with a phone through Micro 4/3 lenses look like. Quite blurry as it turns out, but the colours are kinda nice:
But it was a bit difficult trying various things as it's hard to hold the phone and lens together and try to focus and take a photo at the same time, so I wanted to see if it's possible to fix the focus distance in the Pixel camera, but apparently it is not.
So I searched around on forums a bit and found out about Open Camera for Android. The manual focus functionality didn't come up first which was really frustrating, but I eventually found out that I needed to turn on Camera2 API support:
My Pixel 6 Pro also doesn't have macro lens, so I fixed focus to shortest on the main camera (the telephoto one was quite hopeless with a super long minimum distance) and moved the actual phone until the subjects were sharp(ish). This yielded much nicer photos:
Still nearly not as good as proper macro lens on an actual camera though, but quite all right, and having this capability in the phone that's always with me is of course a huge plus.
I eventually got bored and returned to my computer as I found another interesting article on Escape the algorithm: A brief history of creativity (and power)
I valued creativity since I was a teenager and even ended up studying (and briefly working in) the advertising industry, where everyone is obsessed with it (trying to forget how dirty the actual job of selling more corporate stuff is...). I mean, the biggest ad industry magazine in Hungary is literally called Kreatív (which means creative), but this article brought a completely different perspective:
Pre-archaic cultures not only lacked the language to describe creative acts, but their sociological values were often at odds with the pursuit of originality. In order to access what might be the seeds of modern creativity, we must enter the realm of the gods. Early Mesopotamian theology was ripe with all-powerful female goddesses like Ishtar in Babylonia, who expressed their power through their ability to give birth. Pre-creativity was expressed primarily through procreativity.
How we get from a goddess creating through birth to the (typically male) creative entrepreneurs on the covers of magazines is mainly a story about power. Slowly but surely, male gods climbed the creative ladder, first assisting goddesses in procreation, but eventually overtaking them, as people sought out forms of religious worship that more closely reflected the patriarchal hierarchies of the real world. The invention of writing brought with it an obsession with the power of the kind of immortality that comes from naming something and recording it into history. This was useful to the ancients, as it allowed them to redirect creation from birthing (by women) to breathing and speaking (by men).
The Old Testament story of creation grafted this power from god to man: Adam not only named the animals, but Eve was both born of his ribs and named by his mouth, establishing his dominance over her. Historian Gerda Lerner says that this marks a "counterfactual account of the origin of human life" in which man is literally the woman's mother.
Wow, that last paragraph! But it gets worse:
There are other stories to tell as well. In the twentieth century, creativity became enmeshed with capitalistic productivity, and its study was largely centered around the fields of consumer psychology, organizational management, and product marketing. In other words, the thing being created by creativity is value.
The relationship between creativity and power becomes so strong that it allows for two seemingly contradictory things to occur in parallel: when exercised by the managerial class, it is used to grant capital (think: the deification of the startup founder), and when exercised by the working class, it is used to deny capital (think: the frivolousness ascribed to arts education). Relatedly, the ultimate manifestation of this relationship is the creative logic of social media, wherein users create the content and platforms reap the monetary rewards.
I must admit I am struggling a bit with the urge of doing something "useful" in my free time, so this was a good reminder to stop chasing output (even if it's a non-monetised piece of art).
The essay has a follow-up too, which helps moving beyond this co-opted way of thinking:
Returning to this mindset requires a certain humility, an understanding that simply observing and paying attention to things that already exist and forming connections can be a worthwhile exercise in its own right.
It also involves reevaluating how we assign value to people's time: at what point did we decide that the person who draws a garment on the back of a napkin in five minutes is exercising creativity, but not the person who sits at the sewing machine for years repeatedly making that same garment? As a culture, we have chosen to make a distinction between creativity and the labor of reproduction. But what is reproduction if not the process through which nature creates itself?
There's a lot more in there, do go and read it!
I have to finish my writing though, it's 1:30am and as much as I love the silence of the night, it's time to call it (the night?) a day...